#include "cfm.h"
#include "classifier.h"
#include "coverage.h"
+#include "daemon.h"
#include "dirs.h"
#include "dpif.h"
#include "dynamic-string.h"
COVERAGE_DEFINE(bridge_flush);
COVERAGE_DEFINE(bridge_process_flow);
+COVERAGE_DEFINE(bridge_process_cfm);
+COVERAGE_DEFINE(bridge_process_lacp);
COVERAGE_DEFINE(bridge_reconfigure);
-
-enum lacp_status {
- LACP_STATUS_CURRENT, /* Partner is up to date. */
- LACP_STATUS_EXPIRED, /* Partner is out of date. Attempt to re-sync. */
- LACP_STATUS_DEFAULTED /* Partner information is unknown. */
-};
+COVERAGE_DEFINE(bridge_lacp_update);
struct dst {
uint16_t vlan;
static void dst_set_add(struct dst_set *, const struct dst *);
static void dst_set_free(struct dst_set *);
+enum lacp_status {
+ LACP_CURRENT = 0x01, /* Current State. */
+ LACP_EXPIRED = 0x02, /* Expired State. */
+ LACP_DEFAULTED = 0x04, /* Partner is defaulted. */
+ LACP_ATTACHED = 0x08, /* Attached. Interface may be choosen for flows. */
+};
+
struct iface {
/* These members are always valid. */
struct port *port; /* Containing port. */
const struct ovsrec_interface *cfg;
/* LACP information. */
- enum lacp_status lacp_status; /* LACP state machine status. */
+ enum lacp_status lacp_status; /* LACP status. */
uint16_t lacp_priority; /* LACP port priority. */
struct lacp_info lacp_actor; /* LACP actor information. */
struct lacp_info lacp_partner; /* LACP partner information. */
long long int lacp_tx; /* Next LACP message transmission time. */
long long int lacp_rx; /* Next LACP message receive time. */
- bool lacp_attached; /* Attached to its aggregator? LACP allows
- this link to be chosen for flows. */
};
#define BOND_MASK 0xff
};
enum bond_mode {
- BM_SLB, /* Source Load Balance (Default). */
+ BM_TCP, /* Transport Layer Load Balance. */
+ BM_SLB, /* Source Load Balance. */
BM_AB /* Active Backup. */
};
}
free(managers);
+
+ /* ovs-vswitchd has completed initialization, so allow the process that
+ * forked us to exit successfully. */
+ daemonize_complete();
}
static const char *
? "up" : "down");
error = netdev_get_mtu(iface->netdev, &mtu);
- if (!error) {
+ if (!error && mtu != INT_MAX) {
mtu_64 = mtu;
ovsrec_interface_set_mtu(iface->cfg, &mtu_64, 1);
}
\f
/* Bridge packet processing functions. */
+static bool
+bond_is_tcp_hash(const struct port *port)
+{
+ return port->bond_mode == BM_TCP && port->lacp & LACP_NEGOTIATED;
+}
+
static int
-bond_hash(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
+bond_hash_src(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
{
return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK;
}
+static int bond_hash_tcp(const struct flow *flow, uint16_t vlan)
+{
+ struct flow hash_flow;
+
+ memcpy(&hash_flow, flow, sizeof hash_flow);
+ hash_flow.vlan_tci = 0;
+
+ /* The symmetric quality of this hash function is not required, but
+ * flow_hash_symmetric_l4 already exists, and is sufficient for our
+ * purposes, so we use it out of convenience. */
+ return flow_hash_symmetric_l4(&hash_flow, vlan) & BOND_MASK;
+}
+
static struct bond_entry *
-lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN],
+lookup_bond_entry(const struct port *port, const struct flow *flow,
uint16_t vlan)
{
- assert(port->bond_mode == BM_SLB);
- return &port->bond_hash[bond_hash(mac, vlan)];
+ assert(port->bond_mode != BM_AB);
+
+ if (bond_is_tcp_hash(port)) {
+ return &port->bond_hash[bond_hash_tcp(flow, vlan)];
+ } else {
+ return &port->bond_hash[bond_hash_src(flow->dl_src, vlan)];
+ }
}
static int
if (iface->enabled) {
return i;
} else if (iface->delay_expires < next_delay_expiration
- && (iface->lacp_attached
+ && (iface->lacp_status & LACP_ATTACHED
|| !(port->lacp & LACP_NEGOTIATED))) {
best_down_slave = i;
next_delay_expiration = iface->delay_expires;
}
static bool
-choose_output_iface(const struct port *port, const uint8_t *dl_src,
+choose_output_iface(const struct port *port, const struct flow *flow,
uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags)
{
struct iface *iface;
return false;
}
iface = port->ifaces[port->active_iface];
- } else if (port->bond_mode == BM_SLB){
- struct bond_entry *e = lookup_bond_entry(port, dl_src, vlan);
+ } else {
+ struct bond_entry *e = lookup_bond_entry(port, flow, vlan);
if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
|| !port->ifaces[e->iface_idx]->enabled) {
/* XXX select interface properly. The current interface selection
}
*tags |= e->iface_tag;
iface = port->ifaces[e->iface_idx];
- } else {
- NOT_REACHED();
}
*dp_ifidx = iface->dp_ifidx;
*tags |= iface->tag; /* Currently only used for bonding. */
* They are not required to have synchronized partners because they
* have no partners at all. However, they will only be attached if
* negotiations failed on all interfaces in the bond. */
- up = iface->lacp_attached
+ up = iface->lacp_status & LACP_ATTACHED
&& (iface->lacp_partner.state & LACP_STATE_SYNC
- || iface->lacp_status == LACP_STATUS_DEFAULTED);
+ || iface->lacp_status & LACP_DEFAULTED);
}
return;
}
- if (iface->lacp_status == LACP_STATUS_CURRENT) {
+ if (iface->lacp_status & LACP_CURRENT) {
iface_set_lacp_expired(iface);
}
: in_port->vlan >= 0 ? in_port->vlan
: flow->vlan_tci == 0 ? OFP_VLAN_NONE
: vlan_tci_to_vid(flow->vlan_tci));
- return choose_output_iface(out_port, flow->dl_src, dst->vlan,
+ return choose_output_iface(out_port, flow, dst->vlan,
&dst->dp_ifidx, tags);
}
struct ofpbuf *actions, tag_type *tags,
uint16_t *nf_output_iface, void *br_)
{
- struct iface *iface;
struct bridge *br = br_;
COVERAGE_INC(bridge_process_flow);
+ return process_flow(br, flow, packet, actions, tags, nf_output_iface);
+}
+
+static bool
+bridge_special_ofhook_cb(const struct flow *flow,
+ const struct ofpbuf *packet, void *br_)
+{
+ struct iface *iface;
+ struct bridge *br = br_;
iface = iface_from_dp_ifidx(br, flow->in_port);
if (cfm_should_process_flow(flow)) {
- if (packet && iface->cfm) {
+
+ if (iface && packet && iface->cfm) {
+ COVERAGE_INC(bridge_process_cfm);
cfm_process_heartbeat(iface->cfm, packet);
}
return false;
} else if (flow->dl_type == htons(ETH_TYPE_LACP)) {
- if (packet) {
+
+ if (iface && packet) {
+ COVERAGE_INC(bridge_process_lacp);
lacp_process_packet(packet, iface);
}
return false;
}
- return process_flow(br, flow, packet, actions, tags, nf_output_iface);
+ return true;
}
static void
uint16_t vlan = (flow->vlan_tci
? vlan_tci_to_vid(flow->vlan_tci)
: OFP_VLAN_NONE);
- struct bond_entry *e = lookup_bond_entry(out_port,
- flow->dl_src, vlan);
+ struct bond_entry *e = lookup_bond_entry(out_port, flow, vlan);
e->tx_bytes += n_bytes;
}
}
static struct ofhooks bridge_ofhooks = {
bridge_normal_ofhook_cb,
+ bridge_special_ofhook_cb,
bridge_account_flow_ofhook_cb,
bridge_account_checkpoint_ofhook_cb,
};
return;
}
- iface->lacp_status = LACP_STATUS_CURRENT;
+ iface->lacp_status |= LACP_CURRENT;
+ iface->lacp_status &= ~(LACP_EXPIRED | LACP_DEFAULTED);
iface->lacp_rx = time_msec() + LACP_SLOW_TIME_RX;
iface->lacp_actor.state = iface_get_lacp_state(iface);
size_t i;
struct iface *lead;
struct lacp_info lead_pri;
+ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
port->lacp_need_update = false;
+ COVERAGE_INC(bridge_lacp_update);
if (!port->lacp) {
return;
}
+ VLOG_DBG_RL(&rl, "port %s: re-evaluating LACP link status", port->name);
+
lead = NULL;
for (i = 0; i < port->n_ifaces; i++) {
struct iface *iface = port->ifaces[i];
struct lacp_info pri;
- iface->lacp_attached = true;
+ iface->lacp_status |= LACP_ATTACHED;
ofproto_revalidate(port->bridge->ofproto, iface->tag);
/* Don't allow loopback interfaces to send traffic or lead. */
if (eth_addr_equals(iface->lacp_partner.sysid,
iface->lacp_actor.sysid)) {
- static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
VLOG_WARN_RL(&rl, "iface %s: Loopback detected. Interface is "
"connected to its own bridge", iface->name);
- iface->lacp_attached = false;
+ iface->lacp_status &= ~LACP_ATTACHED;
continue;
}
- if (iface->lacp_status == LACP_STATUS_DEFAULTED) {
+ if (iface->lacp_status & LACP_DEFAULTED) {
continue;
}
for (i = 0; i < port->n_ifaces; i++) {
struct iface *iface = port->ifaces[i];
- if (iface->lacp_status == LACP_STATUS_DEFAULTED
+ if (iface->lacp_status & LACP_DEFAULTED
|| lead->lacp_partner.key != iface->lacp_partner.key
|| !eth_addr_equals(lead->lacp_partner.sysid,
iface->lacp_partner.sysid)) {
- iface->lacp_attached = false;
+ iface->lacp_status &= ~LACP_ATTACHED;
}
}
}
lacp_iface_may_tx(const struct iface *iface)
{
return iface->port->lacp & LACP_ACTIVE
- || iface->lacp_status != LACP_STATUS_DEFAULTED;
+ || iface->lacp_status & (LACP_CURRENT | LACP_EXPIRED);
}
static void
struct iface *iface = port->ifaces[j];
if (time_msec() > iface->lacp_rx) {
- if (iface->lacp_status == LACP_STATUS_CURRENT) {
+ if (iface->lacp_status & LACP_CURRENT) {
iface_set_lacp_expired(iface);
- } else if (iface->lacp_status == LACP_STATUS_EXPIRED) {
+ } else if (iface->lacp_status & LACP_EXPIRED) {
iface_set_lacp_defaulted(iface);
}
}
poll_timer_wait_until(iface->lacp_tx);
}
- if (iface->lacp_status != LACP_STATUS_DEFAULTED) {
+ if (iface->lacp_status & (LACP_CURRENT | LACP_EXPIRED)) {
poll_timer_wait_until(iface->lacp_rx);
}
}
bond_mode_to_string(enum bond_mode bm) {
static char *bm_slb = "balance-slb";
static char *bm_ab = "active-backup";
+ static char *bm_tcp = "balance-tcp";
switch (bm) {
case BM_SLB: return bm_slb;
case BM_AB: return bm_ab;
+ case BM_TCP: return bm_tcp;
}
NOT_REACHED();
struct bond_entry *e;
size_t i;
- assert(port->bond_mode == BM_SLB);
+ assert(port->bond_mode != BM_AB);
/* Sets up 'bals' to describe each of the port's interfaces, sorted in
* descending order of tx_bytes, so that bals[0] represents the most
struct ofpbuf packet;
int error, n_packets, n_errors;
- if (!port->n_ifaces || port->active_iface < 0) {
+ if (!port->n_ifaces || port->active_iface < 0 || bond_is_tcp_hash(port)) {
return;
}
struct flow flow;
int retval;
- if (e->port == port->port_idx
- || !choose_output_iface(port, e->mac, e->vlan, &dp_ifidx, &tags)) {
+ if (e->port == port->port_idx) {
+ continue;
+ }
+
+ compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
+ e->mac);
+ flow_extract(&packet, 0, ODPP_NONE, &flow);
+
+ if (!choose_output_iface(port, &flow, e->vlan, &dp_ifidx, &tags)) {
continue;
}
/* Send packet. */
n_packets++;
- compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
- e->mac);
- flow_extract(&packet, 0, ODPP_NONE, &flow);
retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
&packet);
if (retval) {
bond_mode_to_string(port->bond_mode));
if (port->lacp) {
- ds_put_format(&ds, "\tlacp: %s\n",
+ ds_put_format(&ds, "lacp: %s\n",
port->lacp & LACP_ACTIVE ? "active" : "passive");
} else {
- ds_put_cstr(&ds, "\tlacp: off\n");
+ ds_put_cstr(&ds, "lacp: off\n");
}
+ if (port->bond_mode != BM_AB) {
+ ds_put_format(&ds, "bond-hash-algorithm: %s\n",
+ bond_is_tcp_hash(port) ? "balance-tcp" : "balance-slb");
+ }
+
+
ds_put_format(&ds, "bond-detect-mode: %s\n",
port->miimon ? "miimon" : "carrier");
ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
- if (port->bond_mode == BM_SLB) {
+ if (port->bond_mode != BM_AB) {
ds_put_format(&ds, "next rebalance: %lld ms\n",
port->bond_next_rebalance - time_msec());
}
for (j = 0; j < port->n_ifaces; j++) {
const struct iface *iface = port->ifaces[j];
struct bond_entry *be;
+ struct flow flow;
/* Basic info. */
- ds_put_format(&ds, "slave %s: %s\n",
+ ds_put_format(&ds, "\nslave %s: %s\n",
iface->name, iface->enabled ? "enabled" : "disabled");
if (j == port->active_iface) {
ds_put_cstr(&ds, "\tactive slave\n");
if (port->lacp) {
ds_put_cstr(&ds, "\tstatus: ");
- if (iface->lacp_status == LACP_STATUS_CURRENT) {
+ if (iface->lacp_status & LACP_CURRENT) {
ds_put_cstr(&ds, "current ");
- } else if (iface->lacp_status == LACP_STATUS_EXPIRED) {
+ }
+
+ if (iface->lacp_status & LACP_EXPIRED) {
ds_put_cstr(&ds, "expired ");
- } else {
+ }
+
+ if (iface->lacp_status & LACP_DEFAULTED) {
ds_put_cstr(&ds, "defaulted ");
}
- if (iface->lacp_attached) {
+ if (iface->lacp_status & LACP_ATTACHED) {
ds_put_cstr(&ds, "attached ");
}
ds_put_cstr(&ds, "\tpartner state: ");
ds_put_lacp_state(&ds, iface->lacp_partner.state);
- ds_put_cstr(&ds, "\n\n");
+ ds_put_cstr(&ds, "\n");
}
- if (port->bond_mode != BM_SLB) {
+ if (port->bond_mode == BM_AB) {
continue;
}
/* Hashes. */
+ memset(&flow, 0, sizeof flow);
for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
int hash = be - port->bond_hash;
struct mac_entry *me;
ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
hash, be->tx_bytes / 1024);
+ if (port->bond_mode != BM_SLB) {
+ continue;
+ }
+
/* MACs. */
LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
uint16_t dp_ifidx;
tag_type tags = 0;
- if (bond_hash(me->mac, me->vlan) == hash
+
+ memcpy(flow.dl_src, me->mac, ETH_ADDR_LEN);
+ if (bond_hash_src(me->mac, me->vlan) == hash
&& me->port != port->port_idx
- && choose_output_iface(port, me->mac, me->vlan,
+ && choose_output_iface(port, &flow, me->vlan,
&dp_ifidx, &tags)
&& dp_ifidx == iface->dp_ifidx)
{
if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
== ETH_ADDR_SCAN_COUNT) {
- hash = bond_hash(mac, vlan);
+ hash = bond_hash_src(mac, vlan);
hash_cstr = xasprintf("%u", hash);
unixctl_command_reply(conn, 200, hash_cstr);
port->bond_mode = BM_SLB;
} else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_AB))) {
port->bond_mode = BM_AB;
+ } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_TCP))) {
+ port->bond_mode = BM_TCP;
} else {
port->bond_mode = BM_SLB;
VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s",
bool key_changed;
if (!port->lacp || port->n_ifaces < 1) {
+ for (i = 0; i < port->n_ifaces; i++) {
+ iface_set_lacp_defaulted(port->ifaces[i]);
+ }
return;
}
} else {
size_t i;
- if (port->bond_mode == BM_SLB && !port->bond_hash) {
+ if (port->bond_mode != BM_AB && !port->bond_hash) {
port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
for (i = 0; i <= BOND_MASK; i++) {
struct bond_entry *e = &port->bond_hash[i];
if (port->cfg->bond_fake_iface) {
port->bond_next_fake_iface_update = time_msec();
}
- } else if (port->bond_mode != BM_SLB) {
+ } else if (port->bond_mode == BM_AB) {
free(port->bond_hash);
port->bond_hash = NULL;
}
static void
iface_set_lacp_defaulted(struct iface *iface)
{
- memset(&iface->lacp_partner, 0xff, sizeof iface->lacp_partner);
- iface->lacp_partner.state = 0;
+ memset(&iface->lacp_partner, 0, sizeof iface->lacp_partner);
- iface->lacp_status = LACP_STATUS_DEFAULTED;
+ iface->lacp_status |= LACP_DEFAULTED;
+ iface->lacp_status &= ~(LACP_CURRENT | LACP_EXPIRED);
iface->lacp_tx = 0;
iface->port->lacp_need_update = true;
}
static void
iface_set_lacp_expired(struct iface *iface)
{
- iface->lacp_status = LACP_STATUS_EXPIRED;
+ iface->lacp_status &= ~LACP_CURRENT;
+ iface->lacp_status |= LACP_EXPIRED;
iface->lacp_partner.state |= LACP_STATE_TIME;
iface->lacp_partner.state &= ~LACP_STATE_SYNC;
state |= LACP_STATE_ACT;
}
- if (iface->lacp_status == LACP_STATUS_DEFAULTED) {
- state |= LACP_STATE_DEF;
- } else if (iface->lacp_attached) {
+ if (iface->lacp_status & LACP_ATTACHED) {
state |= LACP_STATE_SYNC;
}
- if (iface->lacp_status == LACP_STATUS_EXPIRED) {
+ if (iface->lacp_status & LACP_DEFAULTED) {
+ state |= LACP_STATE_DEF;
+ }
+
+ if (iface->lacp_status & LACP_EXPIRED) {
state |= LACP_STATE_EXP;
}
mon = iface->cfg->monitor;
if (!mon) {
+ cfm_destroy(iface->cfm);
+ iface->cfm = NULL;
return;
}
projects / openvswitch / blobdiff